一种基于分层复杂性度量的有效的形状图像检索方法

发布时间：2018-05-04 02:09

本文选题：形状描述 + 分层复杂性度量　；参考：《中国科学:信息科学》2017年12期

【摘要】：本文提出了一种基于分层复杂性度量的形状描述方法 (简称HCMD),以应用于一般的形状图像检索任务.该方法属于基于区域的形状描述方法,不需要专门提取形状的边界完成特征抽取任务.本文的主要贡献是提出了一种分层的形状描述框架,基于该框架,形状沿各个方向被迭代地切割成小的区块,每一层级的形状区块被施以各种几何特性的度量,以刻画其形状复杂性.这种分层抽取形状复杂性特征的描述机制使得HCMD具有由粗到细的形状描述能力,能有效地表达形状内部的复杂结构特性.基于HCMD的形状匹配,独立于形状的旋转、缩放、平移和镜像变换,而且计算简单,是一种能有效处理轮廓线形状和区域形状识别的通用方法.用MPEG-7的CE-1轮廓线形状图像库、CE-2区域形状图像库和美国哥伦比亚图像库COIL-20这3个标准测试集对HCMD进行性能评估,并与其他形状描述方法进行了广泛的比较,包括5种基于区域的形状描述子、4种基于点集的形状描述子和两种基于曲线的描述子.实验结果表明HCMD方法在综合考虑检索率、计算效率和一般应用能力指标下,其性能要优于参与比较的各类方法,证明了该方法的有效性.
[Abstract]:In this paper, a method of shape description based on hierarchical complexity metric (HCMDN) is proposed, which can be used in general shape image retrieval. This method belongs to the shape description method based on region and does not need to extract the shape boundary to complete the feature extraction task. The main contribution of this paper is to propose a hierarchical shape description framework, based on which the shape is cut iteratively into small blocks in all directions, and the shape blocks at each level are subjected to various geometric properties. To depict the complexity of its shape. The description mechanism of hierarchical extraction of shape complexity features makes HCMD have the ability to describe the shape from coarse to fine, which can effectively express the complex structural characteristics inside the shape. Shape matching based on HCMD, independent of shape rotation, scaling, translation and image transformation, and simple calculation, is a general method which can effectively deal with contour shape and region shape recognition. The performance of HCMD was evaluated by using the three standard test sets of CE-1 contour shape image library CE-2 and Columbia image library COIL-20 of MPEG-7, and compared with other shape description methods. It includes five region-based shape descriptors, four point-set based shape descriptors and two curvilinear descriptors. The experimental results show that the performance of the HCMD method is better than that of the other methods when considering the index of retrieval rate, computational efficiency and general application ability, and the effectiveness of the method is proved.
【作者单位】：南京财经大学信息工程学院;南京财经大学电子商务省级重点实验室;
【基金】：国家自然科学基金(批准号:61372158) 江苏省自然科学基金(批准号:BK20141487) 江苏省“333工程”高层次人才工程(批准号:BRA2015351) 江苏省科技计划(产学研合作前瞻性联合研究)(批准号:BY2016009-03) 江苏高校优势学科建设工程(PAPD)资助项目
【分类号】：TP391.41

【相似文献】